Notion AI Foresight
Reimagining Notion AI as the AI Operating System for Modern Teams
Notion AI Foresight
Aspiring Product Manager (AI)
Portfolio Case Study
July 2026
Notion AI today is excellent at helping people write, drafting, summarising, and rewriting content inside a page. It is not yet able to tell a team whether its work is actually on track. That gap is costing Notion real revenue: teams that need delivery foresight are paying for a second tool (ClickUp Brain, at $7-9/user/month) or, at the enterprise level, for Glean's enterprise search (roughly $45-50/user/month), more than double what Notion Business costs. This case study proposes Notion Foresight: a new capability that reasons across the docs, comments, and database fields a team already keeps in Notion to surface delivery risk and decision drift before they become a crisis, with every signal traceable back to its source. The MVP is deliberately narrow: one signal type (status language detection), full explainability on every flag, an auto-generated status digest, and unified Business tier pricing with autonomous actions explicitly held back until trust is earned. The core bet: Notion's advantage isn't out-integrating Glean or out-executing ClickUp; it's that the raw signal for foresight is already sitting inside content Notion already owns.
Experience the proposed solution through a working prototype
Read the complete case study with all details in Notion
The Problem
Understanding who we're building for and why it matters
Product question: What would need to be true for Notion to win the AI OS category without out-integrating Glean?
Context & Evidence
Research, data, and observations that informed the product direction
Teams actively paying for a second tool
The delivery-owning team lead pain point
Strategic whitespace validation
The decision: Notion's advantage is reasoning over content it already owns, not out-integrating competitors or out-executing ClickUp on task management.
Goals & Success Metrics
How we define success for users, the product, and the business
Ship one signal type (status-language detection) with high precision; Establish trust through full explainability before adding autonomous actions; Simplify AI pricing into one comprehensible Business tier line item
Surface at-risk projects using signals already present in workspace content; Make every risk signal explainable and independently verifiable in under 10 seconds; Eliminate the recurring manual status-doc workflow for pilot teams
Close the execution intelligence gap that's pushing teams to adopt ClickUp Brain; Defend against Glean's enterprise AI OS positioning by claiming "foresight from owned content"; Create a compelling Business-tier upgrade story beyond writing assistance
Signal dismissal-as-inaccurate rate
Trace panel click-through rate on risk badges
% of eligible projects generating at least one signal within 30 days
Total content creation and collaboration activity must not decline
Zero instances of trace panel surfacing content across permission boundaries
No measurable page-load or workspace performance regression
Constraints
Realistic limitations that shaped the product decisions
Must respect existing page-level permissions
Cannot require new integrations in MVP
Trust must be earned before autonomy
Model cost must fit bundled Business pricing
Having built with LLMs, I know the hardest part isn't the model; it's the reasoning scaffold. Notion's databases already store structured fields alongside unstructured content in one graph. Most competitors have one or the other, rarely both natively linked. That's the architectural advantage.
Options Considered
Different approaches evaluated before deciding on the product direction
F4: Explainable Signal Trace
Every flag links back to its exact source
Trust foundationUser verificationFalse positive management
• Requires UI polish
• Permission-checking overhead
• Must never expose content user can't access
F9: Unified AI Pricing
Fold Foresight into one Business tier line item
Renewal conversation clarityEconomic buyer painPackaging confusion
• Packaging decision, not ML work
• Future features may still need credit metering
F3: Auto-Synthesized Status Digest
Weekly status doc generated from linked content
Manual Friday status doc workflowPersona B core pain
• Complex multi-source synthesis
• Output quality risk
• May be used once as novelty then abandoned
F1: Risk Radar
Surfaces at-risk projects from status/date signals
Core delivery foresight jobClickUp Brain gap
• Genuine ML challenge
• False positive risk
• Status-language detection may misread ambiguous phrasing
F10: Autonomous Re-plan Agent
Proposes/executes schedule changes on detected risk
Ultimate foresight promise
• Most exciting to imagine
• Worst to build first
• Requires trust not yet earned
• Workplace surveillance perception
Trust before autonomy: explainable signals first
Autonomous Re-plan Agent (the exciting feature)
The most exciting feature scored lowest in RICE by design. Foresight signals are worthless if users don't trust them. F4 (Explainable Signal Trace) had to come before F10 (Autonomous Actions).
Prioritization
How decisions were made about what to build and when
RICE (Reach × Impact × Confidence / Effort)
Product Direction & Solution
The product approach chosen and how it addresses the problem
Notion Foresight: a new capability that reasons across workspace docs, comments, and database fields to surface delivery risk and decision drift before they become a crisis
Key Features
Risk Radar: Surfaces at-risk projects from status/date fields plus linked unstructured content
Explainable Signal Trace: Every flag links back to its exact source - clicking a badge opens a trace panel with direct links to source content
Auto-Synthesized Status Digest: Generates weekly status doc live from linked content, replacing manual Friday rebuilds
Unified AI Pricing: Folds Foresight into one coherent Business tier line item, no separate credit metering
User Flow
Notion becomes the AI Operating System for modern teams by turning the content teams already create inside it into foresight about the work itself - not just better writing, but earlier warning.
MVP & Roadmap
What ships first, what comes later, and what we deliberately won't build
- ✓
Risk Radar limited to status-language detection only
- ✓
Explainable Signal Trace shipped simultaneously
- ✓
Manual-only Status Digest generation
- ✓
Foresight included in Business tier at launch
- →
Schedule-conflict signal detection (once status-language precision proven)
- →
Scheduled digest generation with human review before sending
- →
Meeting Note → Risk/Task auto-linker
- →
Foresight ROI Dashboard for renewal conversations
- ✕
Autonomous Re-plan Agent in v1 - trust must be earned first
- ✕
Cross-app integrations in MVP - compete on owned content advantage
- ✕
Admin-configurable risk rules in v1 - ship sensible default first
Schedule-conflict detection in MVP
Ship one signal type (status-language) well with airtight explainability before adding a second. Precision over breadth.
All autonomous actions (F10)
Autonomous re-planning is the most exciting feature to imagine and the worst one to build first. Trust must be earned.
Cross-app integrations (Slack, GitHub)
Compete on owned-content advantage first. Don't race Glean's connector breadth in v1.
Measurement & Experimentation
How we validate the product decisions and measure impact
Foresight-enabled workspaces will show measurable reduction in manually-created status pages versus control, without dropping overall engagement
Three-phase gated rollout with control groups and guardrail metrics at each stage
- ✓
False-positive dismissal rate under internal threshold; zero permission-boundary bugs
Gate to Phase B: precision must be proven internally before external exposure
Phase A: Internal Dogfooding → Phase B: Closed Design-Partner Pilot → Phase C: Phased Business-Tier Rollout
Learnings
What I learned, what surprised me, and what I'd do differently
What I Learned
- •
Category framing was sharp: identified the exact whitespace Notion could own (foresight from owned content) that neither Glean nor ClickUp could replicate
- •
Persona research created a falsifiable hypothesis: Persona B is actively pushing teams to competitors, making her the priority
- •
RICE prioritization put trust (F4) ahead of flashiness (F10), resisting the temptation to over-scope MVP
- •
Experimentation plan had real teeth: kill criteria weren't just box-checking, they were decision gates
What Surprised Me
How much of the competitive threat came from pricing/packaging confusion, not just missing features
What I Would Change
Resisting scope creep even against my own spec: shipping one signal type well teaches more than shipping two signal types with murkier attribution
Run 8-10 structured interviews before writing the PRD instead of synthesizing from public reviews - would have caught blind spots earlier
- ?
What's the right false-positive tolerance before team leads start ignoring badges entirely?
- ?
How should Foresight behave in sparse workspaces—degrade gracefully or suppress signals?
- ?
Could 'risk flagging' be perceived as workplace surveillance even when unintended?
- ?
Would the unit economics (scan cost vs. Business tier pricing) actually work at scale?
PM Reflection
Personal thoughts on the product decisions and process
Every feature, metric, and rollout decision in this case study traces back to a single bet: Notion's advantage is reasoning over content it already owns, not out-integrating Glean or out-executing ClickUp. The MVP is intentionally narrow - one signal type, no autonomy - not because the vision is small, but because trust before autonomy means the roadmap's most ambitious items are only unlocked by proving the boring, trustworthy version works first.